ReID Strong Baseline

Get Started

1. Install dependencies:

   • [pytorch>=0.4]
   • torchvision
   • pytorch-ignite=0.1.2 (Note: V0.2.0 may result in an error)
   • yacs

2. Prepare dataset

   （1）Market1501

   • Download the training and validation set of Market1501
   • Run unzip Market-1501-v15.09.15.zip to unzip the dataset and rename to market1501. The data structure would like:

   data
       market1501 # this folder contains 6 files.
           bounding_box_test/
           bounding_box_train/
           ......
   

   and then you should set the path in all the file in ./test/*.sh about DATASETS.ROOT_DIR

3. Prepare pretrained model if you don't have

   （1）ResNet

   • Download pretrained model resnet50-19c8e357.pth

   • mkdir $ReidStrongBaseline/.cache

   • and then you put it in the path to be ./.cache/*.pth

4. Modify the function of apex package(If you use the apex package after 2021081000, you can ignore it ) add some code in ${PYTHONPATH}/python3.7/site-packages/apex/amp/scaler.py line:307

   if master_grads_combined is None:
       return
   

Train

1. If run the model on the Linux system，you should run the code to convert the format:

   dos2unix test/*.sh
   

2. Market1501, cross entropy loss + triplet loss

cd $ReidStrongBaseline

# training 1p accuracy
bash ./test/train_full_1p.sh --data_path=xxx 
#$data_path for real path to Market1501_datasets

# training 1p performance
bash ./test/train_performance_1p.sh --data_path=xxx 
#$data_path for real path to Market1501_datasets

# training 8p accuracy
bash ./test/train_full_8p.sh --data_path=xxx 
#$data_path for real path to Market1501_datasets

# training 8p performance
bash ./test/train_performance_8p.sh --data_path=xxx 
#$data_path for real path to Market1501_datasets

2. show the prof_demo and inference .

sh demo.sh
sh prof_demo.sh

Test

You can test your model's performance directly by running these commands in .sh files after your custom modification.

# evaluation 1p accuracy
bash ./test/eval_1p.sh --data_path=xxx 
#$data_path for real path to Market1501_datasets

# evaluation 8p accuracy
bash ./test/eval_8p.sh --data_path=xxx 
#$data_path for real path to Market1501_datasets

公网地址说明

代码涉及公网地址参考 public_address_statement.md